A Modified Process Analysis Method and Neural Network Models for Carbon Emissions Assessment in Shield Tunnel Construction

Abstract

This paper proposes a modified process analysis method that combines with the input–output method for carbon emissions assessment in slurry shield tunnel construction. The method was applied to analyze the carbon emissions generated during the construction procedures of a slurry shield tunnel. The results indicate that the carbon emissions from building materials account for the majority of the total emissions, while those from the shield machine and construction procedure are relatively small. In addition, BP and CNN-LSTM neural network models were established to validate the accuracy of the calculation results with model error of 0.1031. Finally, recommendations for reducing carbon emissions in the construction course of slurry shield tunnels are provided.

1. Introduction

The escalating pace of global industrialization and urbanization has led to a significant increase in greenhouse gas emissions. The environmental consequences of global warming pose a severe threat to sustainability [1]. Recognizing the urgency of the situation, the Chinese government has set ambitious targets for peak CO₂ emissions before 2030 and seeks to achieve carbon neutrality by 2060 [2]. It is evident that factors such as industrialization, urbanization, energy consumption intensity, and GDP growth have contributes to the surge in CO₂ emissions [3].

To comprehend the intricate relationship between CO₂ emissions and industrialization and urbanization, researchers such as Xu et al. [4] have explored the trends in different regions of development. They discovered that the trajectory of CO₂ emissions varies across regions undergoing industrialization and urbanization. Notably, the construction of extensive infrastructure, including underground tunnels, has become a significant source of carbon emissions.

Numerous studies have investigated CO₂ emissions in the infrastructure industry. Huang et al. [5] analyzed CO₂ emissions from the construction industry in 40 countries and studied 26 different energy and non-energy sources. Their findings suggested that promoting low carbon building materials, improving construction machinery energy efficiency, and utilizing renewable energy sources could effectively mitigate CO₂ emissions. Another study conducted by Zhang et al. [6] focused on the Chinese construction industry and revealed that CO₂ emissions predominantly originate from the manufacturing of building materials and the operational stages of construction. Similarly, Seo [7] emphasized that material production accounts for the largest share of total carbon emissions, amounting to 93.4%. Furthermore, Chen et al. [8] investigated CO₂ emissions in the Chinese construction industry from 1995 to 2011 and found that indirect emissions constituted more than 95% of the total CO₂ emissions. In terms of construction materials, Choi et al. [9] concluded that utilizing high-strength materials in steel reinforced concrete could lead to significant carbon emissions reductions, while Kim [10] indicated that using reinforced concrete and high-strength concrete structures can reduce CO₂ emissions by approximately 17% compared to other alternatives. Green, low-carbon, recyclable construction materials have also attracted the attention of researchers [11], and these green building materials also have good mechanical properties [12].

The proliferation of tunnel projects is a noteworthy aspect of urbanization construction. These tunnels play a vital role in facilitating convenient transportation and maximizing underground space utilization. By the end of 2020, the lengths of railway tunnels, high-speed railway tunnels, and highway tunnels in China had reached substantial figures of 19,630 km, 6003 km, and 21,999.3 km, respectively. However, tunnel construction significantly contributes to carbon emissions, as the construction sector accounted for 50.6% of China’s total carbon emissions in 2019, with tunnels constituting a significant proportion of these emissions. Xu et al. [13] employed the life cycle assessment (LCA) method to analyze CO₂ emissions in tunnels with varying wall rock conditions. Their study revealed that poor wall rock conditions resulted in an additional 6220–17,010 tons of greenhouse gas emissions during construction, with more than 60% of emissions stemming from materials. Furthermore, Xu et al. [14] developed a linear regression model and employed LCA to analyze the effects of lining design and geological conditions on GHG (greenhouse gas) emissions. Another study conducted by Rodríguez [15] explored CO₂ emissions in tunnels constructed in low- and medium-strength rock masses. It was found that tunnels in medium-strength rock masses emitted 10 t/m, while tunnels with large sections in low-strength rock masses could exhibit a 50% increase in CO₂ emissions. Additionally, Kaewunruen [16] analyzed carbon emissions and energy consumption during tunnel construction and maintenance phases. Their study identified building materials as the main source of carbon emissions, while tunnel maintenance accounted for the majority of energy consumption. Moreover, Guo et al. [17] conducted a study of greenhouse gas emissions in a highway tunnel using the LCA method. They found that each meter of highway tunnel generated 82–113 tons of GHG emissions, with 30.89%, 56.12%, and 12.99% of the total emissions occurring during the construction, operation, and maintenance phases, respectively. Sun and Park [18] also examined CO₂ emissions from tunnel building materials and construction periods, using a tunnel as their case study. Additionally, machine learning techniques have emerged as effective tools for analyzing carbon emissions [19], offering an alternative perspective to the traditional LCA approach.

Despite the significance of tunneling projects and their impact on carbon emissions, limited research has been conducted on shield tunnel construction through different strata. Furthermore, collecting data on carbon emissions and energy consumption in tunneling projects poses challenges due to the complex nature of these projects. Therefore, a modified process analysis method was used to supplement the study of carbon emissions for the whole life cycle of shield tunnels through different strata, including construction equipment, construction materials, construction processes, and especially the manufacturing process of shield machines and the shield boring process. A CNN-LSTM with BPNN neural network was also used to predict the carbon emissions of the shield tunneling part, and the results showed good accuracy. This study will provide a practical reference for reducing carbon emissions during the construction of shield tunnels.

The layout of this paper is as follows. Section 2 introduces various methods for calculating carbon emissions and employs process analysis to evaluate CO₂ emissions. Section 3 presents carbon emissions data related to shield machines, reinforced concrete precast segments, and the tunneling process. Additionally, carbon emissions reduction recommendations are provided, and a novel CNN-LSTM model is employed to compare its results with those obtained using the BPNN model and previous calculations. Finally, Section 4 offers conclusions drawn from the study’s findings and provides suggestions for reducing carbon emissions.

2. Calculation Method and Boundary

2.1. Calculation Method

Currently, the main methods of calculating carbon emissions include LCEA (life cycle energy assessment), LCCO₂A (life cycle carbon emissions assessment), and LCA (life cycle assessment). LCEA is a method used to evaluate the economic aspects of a product or system over its entire life cycle. It involves considering the costs and benefits associated with each stage of the life cycle, including production, use, and disposal. LCCO₂A is a method used to assess the carbon dioxide emissions associated with a product or system throughout its entire life cycle. It involves quantifying the amount of CO₂ emissions released during each stage, including raw material extraction, manufacturing, distribution, use, and disposal. The LCA method includes analyzing the entire process from “new birth” to “death” of buildings and structures. It considers a wide range of environmental factors, such as resource depletion; energy consumption; water usage; emissions to air, water, and soil; and waste generation [20]. LCA considers all stages [21], including raw material extraction, manufacturing, distribution, use, and disposal [22]. For the calculations in this paper, the process analysis method based on LCA is adopted.

Input-output methods and process analysis methods are applications of the LCA approach.

The input–output method adds to the study the perspective of total CO₂ emissions, and it responds to carbon emissions from a completed engineering activity by establishing the relationships of energy with resource consumption and CO₂ emissions.

The process analysis method adds to the study from the perspective of engineering activity, divides an engineering system into subsystems according to the sequence of the activities, calculates carbon emissions from each subsystem separately, and finally totals them to obtain the total carbon emissions. In the case of analyzing and evaluating the effect of a certain activity, it is essential to evaluate not only the environmental problem directly emerging from the research object’s activity but also related effects [23].

To avoid differences in CO₂ calculations caused by different stratum composition, different construction equipment used, and the influences of different surrounding environments during the tunneling of a long-distance tunnel, the tunnel is divided into N different intervals in the calculation, with the intervals containing k rings. Further, to obtain an accurate result for carbon emissions, it is necessary to use the input–output method to revise the process analysis method, adopting the above methods to calculate the same ring carbon emissions in an interval; and according to the calculation results, to revise the carbon emissions factors. Equations (1)–(5) provide the basic formula for the modified process analysis method of calculation [24].

$$E^{RP}={\displaystyle \sum _N{E}_J^{RP}}\left(J=1\cdots N\right)$$

(1)

$$E_J^{RP}=k\cdot E_s^{RP}$$

(2)

$$E_{ES}^I={\eta }_S\cdot E_{ES}^P$$

(3)

$${\eta }_S=\frac{Q_{ES}}{{\displaystyle \sum _j{W}_{Ej}\cdot }{T}_j}$$

(4)

$$E_s^{RP}=e_D\cdot {\displaystyle \sum _i{\beta }_{Di}\cdot W_{Di}}+{\eta }_S\cdot e_E\cdot {\displaystyle \sum _i{\displaystyle \sum _j{W}_{Eij}}}\cdot T_{ij}+{\displaystyle \sum _i{\displaystyle \sum _j{q}_{Mij}\cdot e_{Mj}}}$$

(5)

where $E^{RP}$ is the total carbon emissions of shield tunnel construction calculated by the modified process analysis method, $E_J^{RP}$ is the carbon emissions of a subinterval calculated by the modified process analysis method, $E_s^{RP}$ is the carbon emissions of a standard ring in a subinterval calculated by the modified process analysis method, k is the number of standard rings in a standard interval, $E_{ES}^I$ is energy consumption equipment carbon emissions calculated by the input–output method, $E_{ES}^P$ is energy consumption equipment carbon emissions calculated by the process analysis method, ${\eta }_S$ is a carbon emissions-modified factor of energy consumption equipment, $Q_{ES}$ is the energy consumption of power-consuming equipment obtained by the input–output method, $W_{Ej}$ is the power rating of class j equipment, $T_j$ is the operation time of class j equipment, $e_D$ is the fuel carbon emissions factor, $W_{Di}$ is the number of pieces fuel equipment used in i processes during tunnel construction, ${\beta }_{Di}$ is the fuel consumption in process i during tunnel construction, $e_E$ is the electricity carbon emissions factor, $W_{Eij}$ is the power rating of class j equipment in i processes, $T_{ij}$ is the operation time of class j equipment in i processes, $q_{Mij}$ is the material j consumption in i processes, and $e_{Mj}$ is the carbon emissions factor of material j in i processes.

2.2. CO₂ Emission Source in Tunnel Construction

2.2.1. Shield Machine Parts

The shield machine mainly includes the shield body, cutterhead, propulsion system, screw machine, and segment assembly machine. The shield body is divided into the front shield, middle shield, and tail shield, all of which are tubular cylinders. The cutterhead is located at the front end of the shield and is used to cut the soil. The propulsion system consists of several jacks set inside the middle shield, which is used for shield propulsion and for adjusting the shield attitude. The screw machine transports the soil cut by the cutter backward. The segment assembly machine is located at the end of the shield area and is used to install the lining segment.

The carbon emissions sources of the shield machine can be represented as follows.

1.

Manufacturing of shield machine parts

Including raw material processing, manufacturing, transportation, and processing equipment energy consumption;

2.

Transportation and installation of shield machine parts

The energy consumption of processing equipment and the oil consumption during transportation are included in the calculation;

3.

Transportation and debugging of the shield machine;

4.

Running and maintenance of the shield machine

The energy consumption during the operation of the shield machine and the maintenance equipment are considered;

5.

Demolition and recycling of the shield machine.

Due to the shield machine mechanical equipment production, the mechanical craft is complex, there is a lack of statistical data about CO₂ generated by machine-producing processes, the operating life of the mechanical equipment is longer than their operation hours in the construction period, and the CO₂ emissions from amortization of mechanical equipment are not considered.

In addition, due to the CO₂ emissions of maintenance and recycling constituting a very small proportion of the whole CO₂ emissions in tunnel construction, they are also not considered in the calculation.

The manufacturing process includes:

• The manufacturing of shield machine parts;
• Transportation and installation of shield machine parts.

CO₂ emissions from shield machines manufacturing processes are the main source of the entire carbon emissions in the shield machine life cycle, including CO₂ generated by raw materials used in shield machine manufacturing processes, CO₂ generated by transportation of raw materials, and CO₂ generated by machining of raw materials.

In the manufacturing process, the carbon emissions are the total CO₂ emissions generated by energy and substance consumption during the entire manufacture process. The process consists of the excavation, transportation, machining, and production of raw materials.

In the transportation and installation of shield machine parts, the CO₂ emissions are produced by the energy consumption of assembly instruments, equipment maintenance, and fuel consumption in the transportation process.

Because the construction equipment and transportation instrument use extremely similar energy or fuel, the CO₂ emissions produced by shield part transportation can be determined from the CO₂ emissions produced by construction equipment. Therefore, CO₂ emissions calculations of transportation and construction equipment have the same research boundary.

2.2.2. Tunnel Building Materials

Most of the building materials for tunnels are eventually installed inside the tunnels in the form of reinforced concrete precast segments; therefore, the sources of carbon emissions from building materials are analyzed in terms of reinforced concrete precast segments. To analyze the CO₂ emissions of reinforced concrete precast segments, we use the reinforced concrete precast produced by a factory throughout the year as an example to analyze carbon emissions.

CO₂ is produced during the stages of raw material preparation, transportation, etc. It is necessary to analyze the carbon emissions sources of reinforced concrete precast segments before conducting a study of its carbon emissions [25].

The assumptions of the calculation model are as follows.

• No consideration of climate or environmental impact;
• No consideration of building material recovery factors;
• No consideration of construction equipment depreciation;
• Building material transport distance is a fixed value.
  CO₂ emissions resources for reinforced concrete precast segments can be represented as follows.
• CO₂ generated by non-carbon energy. The main non-carbon energy used in the production of reinforced concrete precast segments is electrical energy. In the case of mainly thermal power generation, it is necessary to calculate the CO₂ emissions generated.
• CO₂ is generated by energy-consuming substances. The dissipation substances used in the production are materials, such as concrete, steel reinforcement, etc. The production of these materials consumes energy and generates CO₂.
• CO₂ is generated by direct combustion of carbonaceous energy. Carbonaceous energy is mainly released by the combustion of fuels, such as in transportation, and maintenance steam is also obtained by burning carbonaceous energy.

2.2.3. Shield Tunneling

According to the study results of energy consumption in broad construction projects [26], the CO₂ emissions of buildings can be divided into construction periods, using period and demolition stage carbon emissions. During the construction period, the main CO₂ emissions come from building materials and construction.

The material CO₂ emissions are mainly the sum of CO₂ emissions from the production and processing of raw materials, including cement, steel, plastics, stone, calcareousness, timber, and so on.

The construction CO₂ emissions are mainly the sum of CO₂ emissions from electricity consumption, water consumption and transportation. A large part of it is related to construction equipment, and the CO₂ emissions calculation of the equipment is mainly performed for the operation period during construction [24].

2.3. Calculation Boundary

There are various raw materials used in the manufacturing process of shield machine equipment and reinforced precast concrete segments. Steel occupies a large proportion of the materials of equipment production. Therefore, it is necessary to determine the calculation boundary of steel. In addition, the calculation boundaries of cement and concrete fabrication must also be determined.

The process of processing and assembling shield machine parts, the production process of reinforce concrete precast segments, and the construction process of shield tunnels use much mechanical equipment. The equipment operations consume large amounts of energy, and due to the different operation processes and equipment types, the types of energy consumed are also different, so it is also necessary to analyze and determine the calculation boundary of various types of mechanical equipment. Moreover, for the equipment used in the tunnel construction process, its service life is much longer than its working time. In this paper, the depreciation maintenance and recycling are not considered, and we only calculate the CO₂ emissions from the operation of the equipment.

A large amount of cement is consumed in the concrete manufacturing process. The production process of cement includes raw material mining, raw material transportation, raw material preparation, clinker calcination, and finished product packaging. Concrete volumes are generally used for engineering quantity statistics. The raw materials used for concrete mixing are cement, water, coarse and fine aggregate, and additives.

In addition, steel consumes much energy in its processing process, generating substantial CO₂ emissions. The process of steel making is roughly as follows: raw material mining, raw material transport, iron and steel making, and rolling and forming of steel.

In addition to carbon emissions from construction materials, shield tunneling also produces some CO₂. The tunnel construction process mainly includes shield excavation, segment assembly, synchronous grouting, slurry circulation ground auxiliary equipment operation, etc.

In summary, the calculation boundary can be determined as follows.

• For the steel, iron and steel making and steel processing are included in the calculation boundary.
• For the cement, raw material preparation, clinker burning, and production packaging are included in the calculation boundary.
• For the sand aggregate, the excavation, mining, and screening process must be considered.
• For the concrete, the mixing process is selected as the calculation content.
• For the material transportation process and construction equipment, the calculation boundary only includes carbon emissions from the working period of the machine equipment, but the carbon emissions from the manufacturing, maintenance, and recycling phases of mechanical equipment are not included in the calculation boundary.

To avoid repeated counting of carbon emissions, carbon emissions from carbon-containing energy sources are considered only in terms of emissions from their use processes, and the carbon emissions from non-carbon energy sources are considered only those caused by their production processes.

3. CO₂ Emissions from Shield Tunnel Construction

3.1. Project Introduction

In this paper, the study is based on the engineering example. The tunnel is about 3390 m long and is designed to be two parallel tunnels with 1695 rings. The burying depth of the land section of the tunnel is 25.2 m to 52.2 m, and the elevation of the section buried underwater is −46.6 m to −54.5 m. The tunnel design is shown in Figure 1.

The outer diameter of the circular tunnel is 14,500 mm, the segment thickness is 600 mm, and the ring width is 2000 mm; moreover, each lining ring is divided into 10 blocks: 7 standard blocks, 2 adjacent blocks, and 1 roof block. The segments are connected by oblique bolts in the longitudinal direction, and the rings are connected by longitudinal bolts. Most of the soil layers through the tunnel are clay silt, silty clay, or silt-sand.

3.2. CO₂ from Tunnel Construction

3.2.1. Shield Machine

We take the diameter of 6510 mm articulated by the earth pressure balance shield machine as an example to analyze and calculate the CO₂ emissions of shield machines in the whole tunnel construction process.

According to relative statistics [27], the crude steel CO₂ emissions factor can be determined as 2.0561 t CO₂/t. The energy consumption of the steel machining process is mainly caused by the manufacturing of the shield carriage, shield shell, and shield mainframe, their process energy consumption parameters are 55.81 kgce/t, 61.85 kgce/t, and 81.75 kgce/t, and their CO₂ emissions factors are 139.15 kg/t, 154.21 kg/t, 203.56 kg/t, respectively.

Since it is difficult to calculate the rate of steel members in shield machine manufacturing processes, and there are no relevant studies, the mainframe takes the same rate as the steel of 87.5%. The carriage is taken as the average of an I-beam of 94.84%. Their CO₂ emissions factors are represented in Figure 2.

The transport and construction equipment used in shield mechanism manufacturing and assembly processes mainly use gasoline, 0# diesel, and electricity as the power energy sources. The energy consumption indices of the corresponding types of machinery and equipment are obtained through actual research and reviews of the literature, and on this basis, the CO₂ emissions factors issued by the IPCC (the Intergovernmental Panel on Climate Change) are considered. Thus, the CO₂ emissions factors of transport and processing equipment are calculated. Among them, machinery and equipment include 16 t bridge cranes, a 30 KVA AC welder, a 75 KVA AC welder, a numerically controlled boring machine (BRD-110R16), a turning center (F50M-3000), and a gantry boring and milling machine (FM-30/80BT); the power of the above equipment is 25 KW, 30 KW, 75 KW, 22 KW, 37 KW, and 75 KW; and the CO₂ emissions factors are 7.84 t/Operating schedule, 9.6 t/Operating schedule, 24 t/Operating schedule, 7.04 t/Operating schedule, 17.76 t/Operating schedule, and 27 t/Operating schedule, respectively. Further, the transport carriers include a 12 t truck and 20 t semi-trailer flat car, their fuel consumption indices are 27.2 L/100 km and 34.2 L/100 km, and the CO₂ emissions factor are 71.75 kg/100 km 90.21 kg/100 km, respectively.

The CO₂ emissions factor of the electricity used in the processing process varies depending on the power generation model. The shield manufacturing CO₂ emissions factor for electricity in the area is 0.817 kg (CO₂)/KWh.

3.2.2. Reinforced Concrete Precast Segments

Taking the segment produced by a factory in one year from the supplier of the reinforced concrete precast segment of the tunnel as an example, the carbon emissions are studied and calculated; moreover, the factory produced 552 rings segments in 2012.

The amount of concrete used in each ring of the tunnel segment is 52 m³ per ring, the amount of reinforcement is 8907.87 kg, and the concrete strength is C60. The CO₂ emissions factors of C60 and steel reinforcement are 447.4 kg CO₂/m³ and 2.3 t CO₂/t, respectively. In addition, the material loss is 5% according to manufacturer data.

Since the concrete is self-made, there is no need to calculate the transportation distance of concrete, so the transportation distance of reinforcement is mainly calculated. The carrier is a 30 t semi-trailer flat car, and the fuel is selected as 0# diesel oil. Its CO₂ emissions factor can be calculated by:

$$EF=\frac{C_{100}\times E{F}_f}{100\times L}$$

(6)

where $EF$ is the CO₂ emissions factor in the transportation process, $C_{100}$ is fuel consumption per 100 km, $E{F}_f$ is the fuel CO₂ emissions factor, and $L$ is the load of transport vehicles.

During the transportation process, the plant used 41.7 t fuel for the transportation of the segment, and the fuel type was 0# diesel, including the fuel used for the forklifts, trucks, etc. The main carrier for road transportation is a 30 t semitrailer flat car, with fuel consumption of 32.3 kg per 100 km, and the average distance is 574 km.

The data related to the prefabricated segments production equipment are shown in Figure 3 and Figure 4.

The electricity CO₂ emissions factor is the same as the value taken above: 0.817 kg CO₂/KW·h.

The CO₂ emissions of reinforced concrete precast segments can be represented by:

$$E_s=E_m+E_t+E_f+E_c$$

(7)

where $E_s$ is the total CO₂ emissions generated in the segment production process, $E_m$ is the CO₂ generated by materials production, $E_t$ is the CO₂ generated from material transportation, $E_f$ is the CO₂ generated by materials processing, and $E_c$ is the CO₂ generated from the segments’ maintenance.

3.2.3. Shield Tunneling

Taking the construction process of a shield tunnel as an example to calculate the CO₂ emissions from construction processes, the CO₂ emissions were calculated according to the power consumption of electrical energy during the operation of the construction equipment, where the power of the equipment at the tunneling site is shown in

Table 1. Equipment power.

Construction Processes	Equipment Names	Power (KW)
Shield excavation	Main drive of shield cutter	3750
	Cooling blower	4.4
	Cooling water	107
	Jacks	200
	Hydraulic system	148
	Segment erector	265
	Synchronous grouting	448
	Air compressor	207
	Ventilation system	60
	Bridge cranes	70
Flue sheet installation	15 t bridge cranes	45
Ground assistant system	20 t bridge cranes	105
	32 t bridge cranes	77
	Tunnel lighting	40
	Repair room	70
	Drainage pump	62
	Ground lighting	54
	Ventilation system	154
Slurry circulation	P1.1(mud pump)	1000
	P2.1(sludge pump)	1000
	P2.2–P2.6 pump (relay pump)	1000
Slurry treatment	Shakespeare system	1241
	Roller screen	15
	Washing and spraying	154
	Industrial pumps	97
	Shear pump	55

.

3.3. Results

Based on the above data, the shield machine CO₂ emissions in the manufacturing phase are indicated in Figure 5.

CO₂ emissions from the transportation process include CO₂ from the transportation of the mainframe and carriage. The calculated transport distance is 320 km. The shield parts’ CO₂ emissions from transportation can be calculated, and the CO₂ emissions of the 12 t truck and 20 t semi-trailer flat car are 10.62 t and 12.75 t, respectively.

The equipment carbon emission factors and the CO₂ emissions from the installation process are represented in Figure 6.

According to the above data, the total CO₂ emissions during the shield manufacturing process are 1673.27 t.

Furthermore, it can be concluded that the CO₂ produced by materials production is 45,940.55 kg per ring, the CO₂ produced by materials transportation is 182.54 kg per ring, and the CO₂ produced by the segment maintenance is 239.72 kg per ring. A total of 412.213 t CO₂ is produced by materials processing. The total CO₂ emission of the segments produced by the factory in one year is 26,304.484 t, and the carbon emissions of the reinforced concrete precast segments for the entire section of the tunnel is 80,771.92 t.

The results of the calculation of CO₂ emissions from the shield tunneling are shown in Figure 7.

Based on the above data, for the total construction of 1695 rings of the tunnel, the total CO₂ emissions are 3540 t.

In the calculation of carbon emissions from the tunnel construction, it can be found that the emissions from steel and concrete account for a large part of the whole carbon emissions, and to improve the efficiency, the use of technical equipment or the use of alternative materials [28] to concrete in some parts of the construction can reduce the carbon emissions of the tunnel construction process, and the carbon emissions from steel can also be reduced [29]. Thus, optimizing the steel manufacturing line, improving the utilization rates of materials, and reasonably adopting different concrete ratios according to the actual conditions of the tunnel can effectively reduce carbon dioxide emissions.

3.4. Neural Network

For neural networks, which are already widely used in the field of carbon emissions analysis, there are often a large number of information features to be processed when analyzing the data. Since the types of influences on neural networks often differ, the ability to extract data features effectively is related to the accuracy of the neural network’s prediction results [30].

Due to the complex parameters that can be extracted by a deep learning algorithm based on CNN and LSTM [31,32], compared with other traditional algorithms, the accuracy of the CNN-LSTM model is relatively higher [33]. The CNN-LSTM model used in this paper consists of a CNN layer and an LSTM layer in turn. The CNN layer in the model is used to withdraw the intricacy features of the parameters that affect carbon emissions, consisting of an input layer, some hidden layers, and an output layer; in addition, the hidden layers consist of a convolutional layer, a batch normalization layer, an Elu layer, and a pooling layer. The parameters pass through the input layer to the convolutional layer for calculation and input to the next layer. Adding the batch normalization layer can avoid the gradient disappearance of the neural network during backpropagation and can greatly accelerate the training speed of the neural network. The convolution layer uses a pooling layer to merge the output of a cluster of neurons from one layer into a single neuron in the next layer; in addition, a regularization layer and a dropout layer are added to the model to avoid overfitting.

Furthermore, the BPNN model is also used to analyze carbon emissions, and to compare the results with the CNN-LSTM, the MAE, MAPE, and RMSE are used to evaluate the two different neural network models by comparing the results obtained from the models and the actual values. The model composition of CNN-LSTM and BPNN are shown in Figure 8 and Figure 9, respectively, the CNN-LSTM model is as described above. The BPNN model consists of an input layer, an implicit layer, and an output layer. The information transmitted by the neurons in the upper layer is passed to the lower layer through the layer-by-layer connection of neurons, the output results of which are obtained by forward propagation and errors by backward propagation after several passes. The model input neurons and output neurons numbers are set as 9 and 1, respectively, and the hidden neuron number according to Equation (8) is set as 12 [34].

$$N=\sqrt{I+O}+R$$

(8)

where N is the number of neurons in the hidden layer, I is the number of neurons in the input layer, O is the number of neurons in the output layer, and R is a random value between 0 and 10.

Compared to traditional RNN, LSTM avoids the gradient disappearance and explosion problems that may occur in traditional RNNs by an input gate, output gate, and forget gate; LSTM controls each gate unit through an activation function that takes values between 0 and 1, and Figure 10 shows the LSTM model. Equations (9)–(11) are the mathematical descriptions of the LSTM model:

$$i_t=\sigma (W_{xi}{x}_t+W_{yi}{y}_{t-1}+b_i)$$

(9)

$$f_t=\sigma (W_{xf}{x}_t+W_{yf}{y}_{t-1}+b_f)$$

(10)

$$o_t=\sigma (W_{xo}{x}_t+W_{yo}{y}_{t-1}+b_o)$$

(11)

where, $W_{xi}$, $W_{xf}$, and $W_{xo}$ are input weights; $W_{yi}$, $W_{yf}$ and $W_{yo}$ are recurrent weights; and $b_i$, $b_f$ and $b_o$ are bias weights.

In the CNN-LSTM model, batch size is a critical hyperparameter that affects the accuracy and training time of the model. Larger batch sizes typically result in higher model accuracy,, but they also require more memory and increase training times. Conversely, smaller batch sizes can reduce memory requirements and training time, but they may lead to model instability and reduced accuracy. To strike a balance between accuracy and training time in this model, we set the batch size to 64. In this model, we set the input dimension to 9, meaning that the model receives input data comprising the nine carbon emission factors mentioned earlier.

Additionally, the output dimension is set to 1, indicating that the model generates a single output, i.e., carbon emissions. We set the dropout rate to 0.15 and the learning rate to 0.005. Given the excellent ability and computational efficiency of the ReLU function in handling nonlinear relationships, we use ReLU as the activation function in the CNN network. Adam is a common optimization algorithm that can adaptively adjust the learning rate and momentum, and we use it as the optimizer in this model.

The number of hidden layers in a model between the input and output layers affects the model’s complexity and training time and has some impact on the model’s accuracy. Too many hidden layers can lead to overfitting and instability. Given the complexity of the model and the size of the training data, we use only one hidden layer in this model.

In

Table 2. Model error.

Model	MAE	MAPE	RMSE
CNN-LSTM	90.1911	0.1031	115.7520
BPNN	110.1934	0.1310	190.1546

, the accuracies of the model results are compared using the MAE, MAPE, and RMSE calculated by the model.

Considering the factors related to carbon emissions during shield tunneling, nine parameters are selected as the variables of the calculation model, including cohesion, internal friction angle, penetration resistance, earth pressure, buried depth, propulsion speed, cutter torque, shield thrust, and operating time of the shield cutter. Since cohesion, internal friction angle, earth pressure, and buried depth all have different effects on the penetration resistance of the penetration resistance, and the cutter torque and shield thrust are related to the soil parameters, in addition, the thrust force has to be readjusted when the soil properties of the different strata change, resulting in a slower shield advance and longer working time. Therefore, the above parameters are given different impact weights to reflect the impacts of these parameters.

Based on the above model, the carbon emissions results of some rings in some sections of the tunnel excavation process were predicted and compared with the carbon emissions calculated based on the process analysis method.

Visibly, the CNN-LSTM model performs significantly better than the BPNN in terms of MAE, MAPE, and RMSE. Compared with the results obtained by the modified process analysis method, the CNN-LSTM has a smaller MAPE value of 0.1031 for the prediction error, while the MAPE of the BPNN reaches 0.1310. As shown in Figure 11, the CNN-LSTM method can obtain an accurate estimate of carbon emissions generated by each ring during shield tunneling, and the prediction of the CNN-LSTM model with appropriate tunneling parameters tends to be close to the actual value. The model error for each sample is also reported in the figure, from which the results obtained by BPNN in sample 22 have a large error, while the CNN model does not show such a large error.

4. Conclusions

This paper uses a modified process analysis to calculate CO₂ emissions from shield tunnels, focusing on CO₂ emissions from the shield machine, the reinforced concrete precast section, and the tunnel construction process. Nine parameters closely related to the operation of the shield machine were also used according to engineering practice, and the selected parameters were input into a neural network model for calculation to determine the carbon emissions.

By calculating the carbon emissions during each phase of tunnel construction, a more intuitive understanding of the main emissions sources can be obtained. The results show that CO₂ emissions from the shield machine, reinforced concrete precast sections and tunnel construction process account for 1673.27 tons, 80,771.92 tons, and 3540 tons, respectively.

Furthermore, in addition to the traditional calculation methods, neural network models also performed well in carbon emission calculations. The proposed neural network models, including CNN-LSTM and BPNN, can obtain relatively accurate estimates of carbon emissions generated during the construction of shield tunnels, which can be applied to the prediction and control of carbon emissions in similar tunnel construction projects. Overall, the use of a combination of traditional calculation methods and neural network models can provide a comprehensive and accurate analysis of carbon emissions in the construction industry.

In the calculation of carbon emissions from tunnel construction, it can be seen that the emissions from steel and concrete account for a large part of the whole carbon emissions. Improving the efficiency of the use of technical equipment and using alternative materials [28] or relatively low-carbon recyclable concrete [11] in some parts of the construction can reduce the carbon emissions during the tunnel construction process, as well as the carbon emissions from steel [29]. Thus, optimizing the steel manufacturing line, improving the utilization rates of materials, and reasonably adopting different concrete ratios according to the actual conditions of the tunnel can effectively reduce carbon dioxide emissions.